Quantitative Research

American Journal of Health Promotion

1-10
Predictors of Supplemental Nutrition ª The Author(s) 2019
Article reuse guidelines:
sagepub.com/journals-permissions
Assistance Program Use at Farmers’ DOI: 10.1177/0890117119854708
journals.sagepub.com/home/ahp
Markets With Monetary Incentive
Programming

Darcy A. Freedman, PhD, MPH1 , David Ngendahimana, PhD1,

En-Jung Shon, PhD2, Kathryn Merritt, MPH3, and Julia Pon, MCP3

Abstract
Purpose: Healthy food incentive program implementation targeting people receiving Supplemental Nutrition
Assistance Program (SNAP) benefits is supported by the federal Food Insecurity Nutrition Incentive (FINI) grant
program. This study examined factors contributing to increased SNAP use at farmers’ markets with an FINI-funded
incentive program.
Design: Implementation evaluation.
Setting: Sixteen states and District of Columbia.
Participants: Two hundred eighty-two FINI-funded farmers’ markets open in 2016.
Measures: Weekly SNAP sales and transactions per 1000 SNAP households in the Zip Code Tabulation Areas around markets.
Analysis: Two-level hierarchical regression modeling.
Results: Most farmers’ markets (53%) had less than 100 SNAP transactions in 2016. Weekly SNAP sales and trans-
actions per 1000 SNAP households were 69.9% and 47.7% higher, respectively, if more than 1 incentive was
available versus 1. Not having paid market staff resulted in declines in these sales (34.3%) and transactions (38.1%)
compared to markets with paid staff. There was a 6.2% and 5.1% increase in SNAP sales and transactions for each
additional produce vendor. Weekly SNAP sales and transactions were about 2 to 3 times higher in rural areas
compared to metropolitan. Clustering of markets within states explained 10% of the variation in weekly SNAP sales
and transactions.
Conclusion: Four implementation factors were identified that may facilitate the reach of SNAP-based monetary incentive
programs at farmers’ markets to maximize reach and impact among SNAP shoppers.

Keywords
food security, nutrition, interventions, behavioral economics, motivation, strategies, farmers market, population health, inter-
ventions, low income, underserved populations, specific populations

Purpose
In the United States, food costs influence dietary decision-
1
making and related behaviors in several ways. First, healthier Department of Population and Quantitative Health Sciences, Case Western
Reserve University, Cleveland, OH, USA
diets tend to be more expensive.1 Americans with higher 2
Department of Family Science and Social Work, Miami University, Oxford,
income can and do spend more money per kilocalorie on foods OH, USA
compared to those with lower income.2 Second, lower income 3
Wholesome Wave, Bridgeport, CT, USA
populations may be less willing to take financial risks to pur-
chase new foods such as a healthier option (ie, fruit or vege- Corresponding Author:
Darcy A. Freedman, Department of Population and Quantitative Health
table) that might be wasted if taste preferences are not met.3 Sciences, Case Western Reserve University, 11000 Cedar Avenue, Room
Third, healthy food choices available for purchase in lower 402-A3, Cleveland, OH 44106, USA.
income neighborhoods are often limited, exacerbating cost Email: [email protected]
2 American Journal of Health Promotion XX(X)

constraints (ie, travel time).4 Taken together, these factors con- participating farmers’ markets. This research examines the
tribute to gradients in diet quality such that each monetary unit incentive program as a lever for promoting SNAP use at farm-
increase in spending on food is associated with corresponding ers’ markets. A key barrier to carrying out this kind of research
increases in conformity to US dietary guidelines.5 in the past has been lack of a centralized system to collect data
The Supplemental Nutrition Assistance Program (SNAP) is related to farmers’ market incentive program implementation.
the largest federal food assistance program in the United States This barrier was addressed in the present study by analyzing
aimed at reducing food cost barriers among low-income popu- data collected through a standardized electronic evaluation
lations. Supplemental Nutrition Assistance Program served platform used at farmers’ markets in 16 US states and the
over 28.8 million households in 2016, the year of the present District of Columbia.24
analysis, providing an average monthly household benefit of
$254.61.6 There is evidence that Americans who qualify for
SNAP benefits and other low-income populations have poorer Methods
diet quality.7-9 Difference in fruit and vegetable consumption
This study was approved as an exempt protocol by the institu-
between high- and low-income Americans is one of the main
tional review board of Case Western Reserve University (pro-
drivers of socioeconomic disparities in overall diet quality.10,11
tocol no. IRB-2014-903).
In the United States, fruits and vegetables often cost more than
other foods.12,13
There is growing interest in population health interventions Description of Study Sample
designed to improve diet quality by reducing food costs for This national-scale implementation evaluation included farm-
low-income populations, such as those receiving SNAP. Fruit ers’ markets participating in FINI-funded projects operating in
and vegetable incentives are an example of this type of inter- 2016 that were using a shared data collection tool for recording
vention approach. Interventions offering a 10% decrease in SNAP payments. Technical support for standardized data col-
prices of fruits and vegetables have been found to increase lection was provided to markets by a national nonprofit,
produce consumption by 14% and reduce body mass index Wholesome Wave.
by 0.04 kg/m2.14 Further evidence suggests SNAP-based fruit All markets included in this study offered monetary incentive
and vegetable incentive programs may reduce disparities in programming to improve the affordability of fruits and vegeta-
cardiovascular disease.15,16 bles available at farmers’ markets for people receiving SNAP
Farmers’ market specific fruit and vegetable incentive pro- benefits. In its simplest version, these SNAP-based incentive
grams are emerging in the United States through support from programs match an expenditure of SNAP benefits with addi-
the federal Food Insecurity Nutrition Incentive (FINI) grant tional funds, which are restricted for fruit and vegetable pur-
program funded by the US Congress in the Agriculture Act chases. In this study, a variety of SNAP-based incentive
of 2014.17 These SNAP-based fruit and vegetable incentives program models were implemented based on local decision-
are an additional food assistance benefit for people receiving making. Incentive program models varied in 3 ways. First, there
SNAP. The goal of SNAP-based incentive programs is to was variability in the incentive matching model. The most com-
increase purchasing of fruits and vegetables while also benefit- mon matching model was a 100% match (eg, spend $1 in SNAP,
ing the local agricultural economy. To achieve population get $1 in incentives). Second, there was variability in the max-
improvements in diet, these SNAP-based monetary incentive imum amount of incentive that could be received per SNAP
programs need to be used by the target population. This is a transaction. Third, markets varied in the number of incentive
challenge because farmers’ markets are an alternative food programs offered. All markets had at least 1 SNAP-based fruit
retail space and barriers, such as lack of technology to accept and vegetable incentive program. In addition, some markets had
SNAP payments, low awareness of farmers markets and incen- other types of incentive programs, not funded by an FINI grant,
tive programs, and irregular hours limit use by low-income that served SNAP customers only or other specified populations.
populations.18 Nevertheless, evidence suggests farmers’ mar- One example of an SNAP-related additional incentive was a
ket use is desirable among some SNAP recipients.19 However, “close the gap” program that supported SNAP recipients at the
farmers’ market use overall tends to be more common among end of the month when federal benefits may have run out. This
populations with higher income.20 Monetary incentive pro- type of program allowed a SNAP recipient to receive the SNAP-
grams are one strategy to address this income gap in farmers’ based incentive by showing their SNAP card without spending
market use. any of their SNAP benefits. Another example is an incentive that
Although there is growth in the number of US farmers’ provided matching funds for people with Medicaid health insur-
markets offering SNAP-based fruit and vegetable incentive ance, a federally funded insurance serving low-income popula-
programs, most research in this area is focused on a small tions in the United States.
number of markets operating similar incentive models in sim-
ilar contexts.21-23 The goal of this research was to examine
heterogeneous models of SNAP-based fruit and vegetable
Measures
incentive programs implemented nationally to identify market, The main outcomes, measured at the market level, were weekly
community, and state factors related to improved SNAP use at SNAP sales and number of SNAP transactions standardized per
Freedman et al. 3

1000 households receiving SNAP within a participating mar- State-level data. Markets were clustered in 16 states and District
ket’s Zip Code Tabulation Areas (ZCTAs).25 More information of Columbia. The states (corresponding number of markets per
about ZCTAs is given subsequently. These variables were com- state) included diverse regions across the United States: Ari-
puted by summing up daily SNAP sales and daily number of zona (2), Connecticut (13), District of Columbia (6), Florida
SNAP transactions and dividing this sum by the number of (17), Georgia (33), Hawaii (2), Maine (36), Missouri (2), New
weeks a market was open in 2016. To standardize the outcomes Hampshire (17), New Jersey (8), Ohio (49), Rhode Island (19),
across diverse contexts, these values were further divided by the Tennessee (4), Virginia (25), Vermont (44), Wisconsin (1), and
number of households receiving SNAP per 1000 households West Virginia (4). State-level variables were related to factors
within a market’s ZCTA.26 Predictor variables for the models hypothesized to influence SNAP use at farmers’ markets. These
included self-reported characteristics about each market as well included the percentage of individuals in the state meeting the
as information about the market’s community and state context. recommended fruit and vegetable intake guidelines as defined
Farmers’ market data were collected via a standardized data by the 2015 to 2020 Dietary Guidelines for Americans29 and
collection platform at all participating markets.24 Staff or dollars of SNAP benefits issued to the state in 2016.30 Addition-
volunteers at each market were trained to enter data into the ally, 5 indicators were included to capture state support for local
platform, and quality control measures were implemented to food systems (eg, number of farmers’ markets per 100 000 res-
address data inconsistencies that were resolved by markets. The idents, state-level food policy council, etc).31
present analysis is focused on data collected from January to
December 2016. Data were exported from the platform in July
2017. A unique market was defined as a market operating at the
Data Analysis
same address during the year. Among all unique markets using All statistical analyses were conducted with R version 3.5.2 (R
the standardized data collection platform in 2016 (N ¼ 407), Core Team, 2018). Baseline characteristics for the study sample
data were limited to markets (a) not enrolled in a randomized were reported using means and standard deviations for continuous
controlled trial organized by Wholesome Wave that was exam- variables that were not skewed and medians and interquartile
ining the impact of different incentive models; (b) open 4 days range (IQR) reported for skewed variables. For categorical vari-
or more in 2016 to exclude holiday markets; (c) had at least 1 ables, frequencies and percentages were calculated. To investi-
SNAP transaction during the year; (d) identified as a farmers’ gate potential market-, community-, and state-level predictors of
market versus a farm stand, mobile market, or community sup- SNAP utilization, we performed separate multilevel linear regres-
ported agriculture; and (e) had SNAP recipients living in the sion analyses to examine characteristics associated with (a) SNAP
ZCTA around the market. This resulted in a total of 282 farm- sales and (b) number of SNAP transaction. For continuous vari-
ers’ markets for the present analysis. ables, annual averages or raw numbers for each market were
entered as independent variables into the model, while for cate-
gorical variables, annual proportions were entered into the model.
Farmers’ market-level variables. Market-level data were entered
Each analysis used a 2-step modeling procedure. In step 1, a series
into the data platform by market staff (eg, the number of weeks
of univariate and correlation analysis were used to examine mar-
a market was open in 2016; whether the market operated on
ket- and community-level characteristics that were associated
weekend day(s) or weekday(s) or both weekend and weekdays;
with the outcome of interest. In step 2, 4 models were fit for each
market manager payment model; number of total vendors at
outcome.32 The 4 models included model 1: with no predictors
market; and number of fruit and vegetable vendors). Additional
and random effect of the intercept (null model); model 2: model 1
dummy variables were assigned for the fruit and vegetable
plus the farmers’ market variables; model 3: model 2 plus state-
incentive maximum benefit (<$10, >$10, or no maximum) and
level variables; and model 4: model 2 plus the state-level variables
number of incentive programs at the market (1 vs 2þ).
with a variance inflation factor below 4.
The model 1 was used to compute intraclass correlation, a
Community-level variables. Community-level variables were cal- measure of variation in the outcomes that is attributable to
culated based on the market zip code entered into the data markets clustering within a state.33 We used Akaike informa-
collection platform. For each market, we defined a community tion criterion (AIC) and Bayesian information criterion (BIC)
using ZCTAs.25 Since each market was associated with a zip to assess model fit to the data. To select a parsimonious model
code rather than a ZCTA, we used a crosswalk to identify among the 5 models, we compared them using ordinary like-
ZCTAs for the markets.27 In this crosswalk, a ZCTA may be lihood ratio tests. The general model for the 2-level mixed-
comprised of one or more zip codes. Using ZCTAs, we linked effect model is specified as follows:
each market to 5-year estimates data on percent of black resi-
X
k
dents using the American Community Survey.26 We designated yij ¼ b0j þ bpx pij þ eij ; eij *Nð0; sij Þ;
each market as being located in a metropolitan, small town, or p¼1
rural area using the primary rural–urban community area
(RUCA) codes.28 Markets with primary RUCA code 1 to 6 X
19
were classified as metropolitan markets, 7 to 9 as small-town b0j ¼ g00 þ g0j Sj þ u0j ; u0j *Nð0; s0j Þ;
markets, and 10 as rural markets. j¼1
4 American Journal of Health Promotion XX(X)

where (ij) denotes the ith market in the jth state such that i ¼ Table 1. Descriptive Statistics of Farmer’s Markets Offering SNAP
1, . . . , 282 and j ¼ 1 . . . , 17 and p represents the pth market Incentives, United States, 2016 (N ¼ 282).
(and community) level hypothesized predictor for the outcome. Results
Note that in this analysis, yij represents the log-transformed
outcome (either SNAP sales or number of SNAP transactions) Farmer’s market operations
for the ith market in the jth state because of large variability in Total annual (2016) SNAP sales, n (%)
the 2 outcomes. Due to this transformation of outcomes, model <$1000 107 (38)
$1001-2000 67 (24)
coefficients were interpreted as percentage increase or
$2001-3000 23 (8)
decrease.34 Specifically, the coefficients were transformed $3001-4000 21 (7)
using the expression that follows.34 For all analyses, statistical >$4000 64 (23)
significance was set at P <.05. Total annual (2016) number of SNAP transactions,
n (%)
100  ðeb  1Þ: <100 150 (53)
101-200 55 (20)
201-300 21 (7)
>300 56 (20)
Results Annual number of transaction weeks, median (IQR) 22 (16)
Number of vendors per week, median (IQR) 15 (19)
Descriptive Statistics Number of produce vendors per week, median (IQR) 5 (5)
Days market is open, n (%)
A total of 282 farmers’ markets were analyzed. These markets Weekday only 109 (39)
had wide variability in terms of market operations. As shown in Weekend and weekday 55 (20)
Table 1, 62% of the markets had total annual SNAP sales below Weekend only 118 (41)
$2000; 23% reported annual SNAP sales above $4000. Fifty- Number of incentives at farmer’s market, n (%)
three percent of the markets had less than 100 SNAP transac- One 202 (72)
Two or more 80 (28)
tions in 2016 and 20% had 300 or more. Among these markets, Spending cap of fruit and vegetable incentive, n (%)
the median number of weeks open across 12 months was 22 $10 105 (37)
(IQR: 16) and the median number of market vendors per week >$10 102 (36)
was 15 (IQR: 19); of these, about 5 were produce vendors. No cap 75 (26)
Many of the markets were open on weekends only (41%) Staffing status of farmer’s market manager, n (%)
or weekdays only (39%). The majority (72%) offered 1 Paid 169 (60)
healthy food incentive program, while 28% offered 2 or Unpaid 70 (25)
Don’t know 43 (15)
more programs. There was variability in the maximum Farmer’s market context
amount of FINI-funded incentive benefits provided per Number of households receiving SNAP in ZCTA of 1158 (1867)
SNAP transaction, which were categorized into 3 groups for farmers’ market, median (IQR)
this analysis, including (a) less than or equal to $10 (37%), Percentage of residents identifying as African 5 (26)
(b) more than $10 (36%), or (c) no limit on the maximum American within ZCTA of farmers’ market,
incentive benefits (26%). median (IQR)
Most of the farmers’ markets (80%) were located within a Geographic area type, n (%)
Metropolitan 226 (80)
metropolitan area with the median size of the ZCTA around the Rural 26 (9)
market of 22 square miles (IQR: 52). In the ZCTA where the Small town 30 (11)
markets were located, the median number of households ZCTA area, in square miles, median (IQR) 22 (52)
receiving SNAP was 1158 (IQR: 1867) and residents identify- State level (16 states and Washington, DC)
ing as African American was 5% (IQR: 25%), indicating varia- Percentage of residents meeting recommended fruit 12.3 (2)
bility in the sociodemographic context of the participating intake, mean (SD)
farmers’ markets. About half of the markets were located in Percentage of residents meeting recommended 9.2 (2)
vegetable intake, mean (SD)
4 states: Ohio (17%), Vermont (16%), Maine (13%), and Number of farmers’ markets per 100 000 residents, 4.2 (5)
Georgia (12%). On average, for the 16 states and District of median (IQR)
Columbia, 12% and 9% of the residents met the recommended Percentage of farmers’ markets accepting WIC, 27.6 (30)
fruit and vegetable intake, respectively. Median annual SNAP median (IQR)
issuance to the state was $923 (IQR: $1130) million per state. State-level food policy council, N (%) 11 (65)
Number of local food policy councils (mean, SD) 4 (6)
Number of food hubs, mean (SD) 3.9 (3)
Model Selection SNAP issuance in million dollars, median (IQR) 923 (1130)

The following results are related to 2 main outcomes including Abbreviations: IQR, interquartile range; M, mean; SD, standard deviation;
SNAP, Supplemental Nutrition Assistance Program; WIC, Special Supplemen-
weekly SNAP sales and number of SNAP transactions per 1000 tal Nutrition Program for Women, Infants, and Children; ZCTA, ZIP Code
households receiving SNAP within a participating market’s Tabulation Areas.
Freedman et al. 5

Table 2. Estimates From 2-level Mixed-Effect Model of Weekly SNAP Sales ($) in 2016 Per 1000 Households Receiving SNAP Benefits Within a
Farmers’ Market ZCTA (N ¼ 282 Farmers’ Markets).

Model 1a Model 2a Model 3a Model 4a

Fixed Effect Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE)

Farmer’s market operations

Intercept 4.40 (0.16) 3.69 (0.31)b 1.81 (9.08) 3.50 (0.46)
Spending cap of fruit and vegetable incentive (ref ¼ $10 or less)
More than $10 0.21 (0.26) 0.59 (0.28) 0.41 (0.27)
No cap 0.20 (0.27) 0.49 (0.29) 0.41 (0.28)
Number of incentives of farmer’s market
Two or more (ref ¼ 1) 0.46 (0.21)b 0.60 (0.21)b 0.53 (0.21)b
Market open days (ref ¼ weekday only)
Weekend and weekday 0.09 (0.23) 0.06 (0.23) 0.06 (0.23)
Weekend only 0.002 (0.18) 0.14 (0.18) 0.14 (0.18)
Number of vendors per week 0.02 (0.01)b 0.02 (0.008)b 0.02 (0.01)b
Number of produce vendors per week 0.06 (0.02)b 0.05 (0.02) 0.06 (0.02)b
Staffing status of farmers market manager (ref ¼ paid)
Unpaid 0.43 (0.21) 0.39 (0.21) 0.42 (0.20)
Farmer’s market context
Percentage of residents identifying as African American within markets’ 0.02 (0.01) 0.02 (0.01) 0.02 (0.01)b
ZCTA
Geographic area type (ref ¼ metropolitan)
Rural 1.41 (0.31) 1.36 (0.30) 1.42 (0.30)b
Small town 0.25 (0.29) 0.34 (0.29) 0.25 (0.28)
State context
Percentage of residents meeting recommended fruit intake 0.57 (0.40)
Percentage of residents meeting recommended vegetable intake 0.85 (0.50)
SNAP issuance in million dollars 0.09 (0.41)
Number of FM per 100 000 residence 0.23 (0.14)
Percentage of farmers accepting WIC Program Vouchers 0.01 (0.02)
State-level food policy council 0.28 (0.73) 0.07 (0.48)
Number of local food policy councils 0.06 (0.07) 0.01 (0.05)
Number of food hubs 0.03 (0.09)
Model summary
Random effects
Level 1 1.45b
Level 2 intercept (state mean) 0.47b
Model fit
AIC (less the better) 1032.16 827.34 819.49 798.01
BIC (less the better) 1043.08 876.01 895.41 853.23

Abbreviations: AIC, Akaike information criterion; BIC, Bayesian information criterion; SE, standard error; SNAP, Supplemental Nutrition Assistance Program;
WIC, Special Supplemental Nutrition Program for Women, Infants, and Children; ZCTA, Zip Code Tabulation Area; HLM, Hierarchical Linear Modeling.
a
Model 1: HLM with only random intercept (null model); model 2: model 1 plus the farmers’ market variables; model 3: model 2 plus state-level variables; and
model 4: model 2 plus the state-level variables with a variance inflation factor below 4.
b
P < .05.

ZCTA. Henceforth, these are referred to as Weekly SNAP Sales Table 4 provides results of the 2-level mixed-effect models
and Weekly SNAP Transactions. predicting Weekly SNAP Sales. There were 6 statistically sig-
As shown in Table 2, the Weekly SNAP Sales model nificant predictors of Weekly SNAP Sales at the participating
(model 4) with market-level variables and 2 state-level farmers’ markets. Weekly SNAP Sales at markets were 69.9%
variables had the least AIC and BIC (798.01 and 853.23, higher if 2 or more incentive programs were available versus
respectively) indicating best fit. As shown in Table 3, for only 1. Weekly SNAP Sales increased by 2% for every addi-
Weekly SNAP Transactions, model 4 provided the best fit tional vendor per week and 6.2% for every additional produce
with the least AIC and BIC (761.51 and 816.73, respec- vendor. There was a 34.3% decrease in Weekly SNAP Sales at
tively). The ICC for both Weekly SNAP Sales and Weekly farmers’ markets with an unpaid volunteer serving as the farm-
SNAP Transactions was 0.10, which was statistically signif- ers’ market manager compared to a paid staff member serving
icant at P < .05. Thus, 10% of the variation of Weekly this role. Weekly SNAP Sales decreased 2% for each percent-
SNAP Sales and Weekly SNAP Transactions were due to age increase in the number of households identifying as African
clustering of markets within states. Americans in the ZCTA where the farmers’ market was
6 American Journal of Health Promotion XX(X)

Table 3. Estimates From 2-Level Mixed-Effect Model of Weekly SNAP Transactions in 2016 per 1000 Households Receiving SNAP Benefits
Within a Farmers’ Market ZCTA (N ¼ 282 Farmers’ Markets).

Model 1a Model 2a Model 3a Model 4a

Fixed Effect Estimate (SE) Estimate (SE) Estimate (SE) Estimate (SE)

Farmer’s market operations

Intercept 1.64 (0.14) 1.20 (0.28) 1.68 (6.40) 0.85 (0.44)
Spending cap of fruit and vegetable incentive (ref ¼ $10 or less)
More than $10 0.04 (0.25) 0.48 (0.25) 0.27 (0.25)
No cap 0.007 (0.25) 0.36 (0.26) 0.22 (0.26)
Number of incentives of farmer’s market
Two or more (ref ¼ one) 0.33 (0.20) 0.46 (0.19)b 0.39 (0.19)b
Market open days (ref ¼ weekday only)
Weekend and weekday 0.02 (0.21) 0.15 (0.21) 0.15 (0.21)
Weekend only 0.05 (0.17) 0.10 (0.17) 0.09 (0.16)
Number of vendors per week 0.02 (0.007)b 0.01 (0.007) 0.01 (0.01)
Number of produce vendors per week 0.05 (0.02)b 0.05 (0.02)b 0.05 (0.02)b
Staffing status of farmers market manager (ref ¼ paid)
Unpaid 0.50 (0.20) 0.45 (0.19)b 0.48 (0.19)b
Farmer’s market context
Percentage of residents identifying as African American within markets’ 0.02 (0.004)b 0.01 (0.005) 0.01 (0.005)
ZCTA
Geographic area type (ref ¼ metropolitan)
Rural 1.12 (0.29)b 1.04 (0.28)b 1.13 (0.27)b
Small town 0.27 (0.27) 0.39 (0.26) 0.30 (0.26)
State context
Percentage of residents meeting recommended fruit intake 0.60 (0.31)
Percentage of residents meeting recommended vegetable intake 0.86 (0.39)
SNAP issuance in million dollars 0.12 (0.29)
Number of FM per 100 000 0.24 (0.10)
Percentage of farmers accepting WIC program vouchers 0.02 (0.01)
State food policy council 0.40 (0.55) 0.14 (0.45)
Number of local food policy councils 0.06 (0.05) 0.02 (0.04)
Number of food hubs 0.04 (0.06)
Model summary
Random effects
Level 1 1.29b
Level 2 intercept (state mean) 0.42b
Model fit
AIC (less the better) 966.73 795.74 782.73 761.51
BIC (less the better) 977.66 844.41 858.65 816.73

Abbreviations: AIC, Akaike information criterion; BIC, Bayesian information criterion; SNAP, Supplemental Nutrition Assistance Program; ZCTA, Zip Code
Tabulation Area; WIC, Special Supplemental Nutrition Program for Women, Infants, and Children, SE; standard error; HLM, Hierarchical Linear Modeling.
a
Model 1: HLM with only random intercept (null model); model 2: model 1 plus the farmers’ market variables; model 3: model 2 plus state-level variables; and
model 4: model 2 plus the state-level variables with a variance inflation factor below 4.
b
P < .05.

located. Markets located in rural areas had 3 times as many about 2 times as many Weekly SNAP Transactions as those
Weekly SNAP Sales as those located in metropolitan areas. located in metropolitan areas.
Table 4 also highlights results of the 2-level mixed-effect
models predicting Weekly SNAP Transactions. There were 4
statistically significant predictors of Weekly SNAP Transac- Discussion
tions. All 4 also were statistically significant predictors of Evidence is growing in support of monetary incentive pro-
Weekly SNAP Sales. Weekly SNAP Transactions at markets grams as a health promotion strategy to improve fruit and
were 47.7% higher if 2 or more incentive programs were avail- vegetable consumption among low-income populations. Incen-
able versus only 1. Weekly SNAP Transactions increased 5.1% tive intervention studies have focused on different food retail
for every additional produce vendor. There was a 38.1% environments such as supermarkets, worksite cafeterias, and
decrease in Weekly SNAP Transactions at markets with an farmers’ markets and found improvements in the sale of fruits
unpaid volunteer serving as the farmers’ market manager com- and vegetables at participating retailers and/or improvement in
pared to a paid staff member. Markets located in rural areas had dietary intake or body mass index among participants using the
Freedman et al. 7

Table 4. Estimates From 2-Level Mixed-Effect Model Predicting Weekly SNAP Sales and Number of Transactions in 2016 per 1000 Households
Receiving SNAP Benefits Within a Farmers’ market ZCTA (N ¼ 282 Farmers’ Markets).

Weekly SNAP Sales ($) Modela Weekly SNAP Transactions Modela

Fixed Effect Coefficient (SE) Percent Change Coefficient (SE) Percent Change

Market operations
Spending cap of fruit and vegetable incentive (ref ¼ $10 or less)
More than $10 0.41 (0.27) 50.7 0.27 (0.25) 31
No cap 0.41 (0.28) 50.7 0.22 (0.26) 24.6
Number of incentives of farmer’s market
Two or more (ref ¼ 1) 0.53 (0.21)b 69.9 0.39 (0.19)b 47.7
Market open days (ref ¼ weekday only)
Weekend and weekday 0.06 (0.23) 6.2 0.15 (0.21) 16.2
Weekend only 0.14 (0.18) 15 0.09 (0.16) 9.2
Number of vendors per week 0.02 (0.01)b 2 0.01 (0.007) 1
Number of produce vendors per week 0.06 (0.02)b 6.2 0.05 (0.02)b 5.1
Staffing status of farmers market manager (ref ¼ paid)
Unpaid 0.42 (0.20)b 34.3 0.48 (0.19)b 38.1
Farmer’s market context
Percentage of residents identifying as African American within 0.02 (0.005)b 2 0.01 (0.005) 1
markets’ ZCTA
Geographic area type (ref ¼ metropolitan)
Rural 1.42 (0.30)b 313.7 1.13 (0.27)b 209.6
Small town 0.25 (0.28) 22.1 0.30 (0.26) 25.9
State context
State-level food policy council 0.07 (0.48) 6.8 0.14 (0.45) 15
Number of local food policy councils 0.01 (0.05) 1 0.02 (0.04) 2
Model summary
Random Effects Standard Deviation Standard Deviation
Level 1 1.45b 1.29b
Level 2 intercept (state mean) 0.47b 0.42b

Abbreviations: SE, standard error; SNAP, Supplemental Nutrition Assistance Program; ZCTA, Zip Code Tabulation Area.
a
Intraclass correlation (ICC) ¼ 0.10.
b
P < .05.

incentives.35-39 Moreover, people receiving SNAP benefits incentive program. On average, there were 167 SNAP house-
report interest in interventions that incentivize healthy foods holds residing near the markets in rural settings versus 1783 in
such as fruits and vegetables.40 metropolitan settings.26 There were an average of 34 weekly
Given this evidence, results of our study add to the field by SNAP transactions per 1000 SNAP households in rural settings
identifying implementation factors to maximize the reach of versus 21 in metropolitan settings. The overall number of peo-
monetary incentive programs among SNAP consumers shop- ple reached was higher in urban areas (a function of population
ping at farmers’ markets. This is important because results of density); however, the rate of utilization of the SNAP incentive
the present study reveal the overall reach of the FINI-funded program was significantly higher in rural areas. This is in con-
SNAP-based incentive programs offered at participating farm- trast to prior research that found barriers to using farmers’
ers’ markets was low. More than half of the markets included in markets in rural settings.41 Higher rates of SNAP use at the
this study had less than 100 SNAP transactions in 2016. Find- farmers’ markets located in rural areas may be due to the fact
ings highlight several implementation factors that may facil- that there is higher farmers’ market density in metropolitan
itate the reach of SNAP-based fruit and vegetable incentive areas, which provides greater choice to SNAP consumers but
programs at farmers’ markets. distributes the consumer base among multiple markets. This may
First, findings highlight that location matters. Farmers’ mar- result in lower rates of SNAP use at some of the markets in
kets in rural communities had 3 times more weekly SNAP sales metropolitan areas. Additionally, findings reveal weekly SNAP
per 1000 SNAP households in the area surrounding the farm- sales per 1000 SNAP households in the area surrounding parti-
ers’ market compared to markets in metropolitan areas. Weekly cipating farmers’ markets were about 2% lower for each per-
SNAP transactions per 1000 SNAP households were 2 times centage increase in African American residents living in the area
higher at rural markets. Although fewer SNAP households are near the market. Together, these findings highlight the benefits
located near markets in rural settings, these households of targeting urban African American populations in future
were more likely to be reached by the farmers’ market SNAP SNAP-based incentive program implementation and outreach.
8 American Journal of Health Promotion XX(X)

Second, there was evidence that offering more than 1 incen- of these interventions among people receiving SNAP. Further-
tive program was strongly related to improvements in SNAP more, FINI implementation should target and tailor outreach
use. Multiple incentive programs may motivate cost-conscious within specific geographic areas with high rates of households
customers to frequent a farmers’ market because there will be receiving SNAP (eg, StrikeForce zones in rural areas, public
several opportunities to extend financial resources for food housing communities in urban areas).
within 1 shopping visit. Multiple incentive programs also pro- From a programmatic perspective, findings provide a
vide opportunities for shoppers to take advantage of different framework for conceptualizing the core elements of
subsidies as resources ebb and flow across the monthly SNAP SNAP-based fruit and vegetable incentive program imple-
distribution cycle. mentation at farmers’ markets. Study findings suggest core
Third, findings support paying farmers’ market manage- elements include having (1) more than 1 incentive program,
ment staff to operate incentive programs. Paid market staff may (2) paid farmers’ market staff responsible for incentive pro-
be more likely to be a regular presence at the market and have gramming, (3) high numbers of produce vendors available
more time to conduct marketing and outreach for the market on each market day, and (4) outreach to garner more cus-
and incentive program outside of market hours. Consistency of tomers overall combined with targeted efforts to promote
market staff fosters relationships with customers who may be farmers’ market use among African American SNAP reci-
more likely to return because of this social connection.42 pients living in urban communities.
Fourth, results highlight the importance of increasing the There are strengths and limitations to this research. A key
number of produce vendors selling at the market. Strategic strength is the sample size, including diversity in market mod-
growth of market size will require targeted efforts to expand els, locations, and operations. To our knowledge, this study
the number of fruit and vegetable vendors who may be in short represents the largest sample of farmers’ markets included in
supply in some communities. To promote inclusivity, this type a study focused on incentive programming. Another strength is
of expansion may focus on including vendors with similar
sociodemographic characteristics of the targeted SNAP popu-
lation.18 Some markets have diversified their vendor base
through efforts to connect with immigrant and refugee farmers SO WHAT?
who often grow foods preferred by different ethnic groups.
Together, these implementation factors address transac-
What is already known on this topic?
tional costs that influence economic decision-making. Transac-
tional costs include all of the costs beyond the price of the food  Healthy food incentive programs are effective
that influence a decision to purchase a good.43 These are espe- strategies to promote dietary improvements.
cially important since monetary incentive interventions target  There is growing interest in implementing these
the price of the food but not necessarily these other transac- types of programs at farmers’ markets.
tional costs. At a farmers’ market, these may include social  Strategies are needed to maximize the reach of
costs of interacting in a new space that may not feel culturally farmers’ market incentive programs among peo-
inclusive, psychic costs of engaging in a food shopping beha- ple receiving SNAP.
vior that may not be normative, informational costs to deter-
mine which foods to purchase and how to prepare them, What does this article add?
infrastructure costs such as access to refrigeration or storage
to effectively manage the foods at home, transportation and  This is the largest study evaluating real-world
time costs to get to the market, and the additional costs needed implementation of incentive programming at
to complete food purchasing if all food needs cannot be met at farmers’ markets.
the farmers’ market.  Results provide evidence of implementation fac-
Results have implications for policy and practice. From a tors related to increased use of SNAP benefits at
policy perspective, findings offer guidance for enhancing the farmers’ markets offering incentive programs.
federal FINI grant program to maximize reach among SNAP
recipients. This is particularly timely because the Agriculture What are the implications for health promotion
Improvement Act of 2018 (ie, US Farm Bill) includes expan- practice or research?
sion of the FINI grant program to $250 million over 5 years and
establishes the program as a permanent part of future farm  Findings have implications for federal nutrition
bills.44 Findings from this research indicate there is a benefit policy implementation by identifying programma-
to providing financial support to farmers’ markets to cover the tic components that are likely to increase the
costs of hiring staff to operate fruit and vegetable incentive impact of healthy food incentive programs.
programming. Additionally, policies that support incentivizing  Findings provide an evidence-based framework for
multiple forms of federal subsidies such as Medicaid or Special expansion of SNAP-based incentive programming
Supplemental Nutrition Program for Women, Infants, and Chil- to maximize reach among people receiving SNAP.
dren within the farmers’ market context may further the reach
Freedman et al. 9

the analytic approach that allowed for examination of multiple Cooperative Agreement Number 1U48DP005013 from the Centers for
factors influencing SNAP utilization while taking into account Disease Control and Prevention (CDC) and was supported by grant
nesting of farmers’ markets within states. Weaknesses include number 2015-70018-23350 from the US Department of Agriculture
risk of data entry errors from market staff or volunteers. There (USDA). The findings and conclusions in this manuscript are those of
the authors and do not necessarily represent the official position of the
may be other important factors influencing SNAP utilization,
CDC or USDA. The CDC and USDA had no role in the design,
such as marketing and communication approaches that were
analysis, or writing of this article.
not captured across sites. Another limitation is the fact that this
study lacks data on non-FINI-funded farmers’ markets that
may be located near the markets included in this study, which ORCID iD
could impact SNAP utilization rates per market. Finally, the Darcy A. Freedman https://orcid.org/0000-0002-0521-9621
markets included in this study are different from other farmers’
market across the United States because they received funding
from an FINI grant to support SNAP incentive programming. References
Findings may not be generalizable to non-FINI-funded farm- 1. Darmon N, Drewnowski A. Contribution of food prices and diet
ers’ markets. cost to socioeconomic disparities in diet quality and health: a
In conclusion, this study extends the science of healthy food systematic review and analysis. Nutr Rev. 2015;73(10):643-660.
incentive programming by examining implementation factors doi:10.1093/nutrit/nuv027.
that can be modified to maximize impact among SNAP con- 2. Rehm CD, Monsivais P, Drewnowski A. Relation between diet
sumers shopping at farmers’ markets. This research builds on cost and healthy eating index 2010 scores among adults in the
the growing body of evidence showing support for SNAP- United States, 2007–2010. Prev Med (Baltim). 2015;73:70-75.
based monetary incentives as a strategy to improve dietary doi:10.1016/j.ypmed.2015.01.019.
patterns among low-income populations. Our analysis of the 3. Daniel C. Economic constraints on taste formation and the true
real-world implementation of farmers’ market fruit and vege- cost of healthy eating. Soc Sci Med. 2016;148:34-41. doi:10.1016/
table incentive programs operating throughout the United J.SOCSCIMED.2015.11.025.
States reveals strategic targets to guide future policy and pro- 4. Dutko P, Ver Ploeg M, Farrigan T. Characteristics and Influential
grammatic initiatives. Factors of Food Deserts. In: Service ER, ed. Washington, DC: US
Department of Agriculture; 2012.
Authors’ Note 5. Aggarwal A, Monsivais P, Drewnowski A. Nutrient intakes
D. Freedman and K. Merritt conceptualized this manuscript. K. Mer- linked to better health outcomes are associated with higher diet
ritt and J. Pon supported data collection. D. Ngendahimana and E. costs in the US. PLoS One. 2012;7(5):e37533. doi:10.1371/jour-
Shon conducted analysis. All authors were involved in manuscript nal.pone.0037533.
writing and approval. This study was approved as an exempt protocol 6. Food and Nutrition Service. Supplemental Nutrition Assistance
by the Case Western Reserve University institutional review board Program: FY16 Through FY19 National View Summary.
(protocol no. IRB-2014-903). Washington, DC: U.S. Department of Agriculture; 2019.
7. Hiza HAB, Casavale KO, Guenther PM, Davis CA. Diet quality
Acknowledgments of americans differs by age, sex, race/ethnicity, income, and edu-
The authors thank the farmers’ market staff and volunteers as well as cation level. J Acad Nutr Diet. 2013;113(2):297-306. doi:10.
Wholesome Wave staff for their support with data collection. Addi- 1016/J.JAND.2012.08.011.
tionally, we honor the wisdom of the late Gus Schumacher who pro- 8. Wang DD, Leung CW, Li Y, et al. Trends in dietary quality
vided advice and encouragement for the development and among adults in the United States, 1999 through 2010. JAMA
implementation of this national fruit and vegetable incentive program.
Intern Med. 2014;174(10):1587-1595. doi:10.1001/jamain-
The authors are grateful to the reviewers for their thoughtful feedback
ternmed.2014.3422.
on earlier versions of this article.
9. Leung CW, Ding EL, Catalano PJ, Villamor E, Rimm EB, Willett
Declaration of Conflicting Interests WC. Dietary intake and dietary quality of low-income adults in the
supplemental nutrition assistance program 1–3. Am J Clin Nutr. 2012;
The author(s) declared the following potential conflicts of interest
96(5):977-988. doi:10.3945/ajcn.112.040014.INTRODUCTION.
with respect to the research, authorship, and/or publication of this
article: D. Freedman invented software used for data collection in this 10. Guenther PM, Juan WY, Lino M, Lucas R. Diet Quality of Low-
study and this software is being commercialized by an outside entity. Income and Higher Income Americans in 2003-04 as Measured
Case Western Reserve University is the owner of the invention being by the Healthy Eating Index-2005. Washington, DC: U.S. Depart-
commercialized by this outside entity, and as such, could gain ment of Agriculture; 2008.
royalties. 11. Storey M, Anderson P. Income and race/ethnicity influence diet-
ary fiber intake and vegetable consumption. Nutr Res. 2014;
Funding 34(10):844-850. doi:10.1016/J.NUTRES.2014.08.016
The author(s) disclosed receipt of the following financial support for 12. Drewnowski A, Darmon N, Briend A. Replacing fats and sweets
the research, authorship, and/or publication of this article: This with vegetables and fruits—a question of cost. Am J Public
research is a product of a Prevention Research Center supported by Health. 2004;94(9):1555-1559. doi:10.2105/AJPH.94.9.1555.
10 American Journal of Health Promotion XX(X)

13. Drewnowski A. The cost of US foods as related to their nutritive 29. Lee-Kwan SH, Moore LV, Blanck HM, Harris DM, Galuska D.
value. Am J Clin Nutr. 2010;92(5):1181-1188. doi:10.3945/ajcn. Disparities in state-specific adult fruit and vegetable consumption
2010.29300. – United States, 2015. Morb Mortal Wkly Rep. 2017;66(45):
14. Afshin A, Peñalvo JL, Del Gobbo L, et al. The prospective impact 1241-1247.
of food pricing on improving dietary consumption: a systematic 30. Supplemental Nutrition Assistance Program State Activity
review and meta-analysis. PLoS One. 2017;12(3):e0172277. doi: Report. 2016. https://fns-prod.azureedge.net/sites/default/files/
10.1371/journal.pone.0172277. snap/FY16-State-Activity-Report.pdf. Accessed May 2, 2018.
15. Pearson-Stuttard J, Bandosz P, Rehm CD, et al. Reducing US 31. Centers for Disease Control and Prevention. State Indicator
cardiovascular disease burden and disparities through national Report on Fruits and Vegetables. Atlanta, GA: US Department
and targeted dietary policies: a modelling study. PLoS Med. of Health and Human Services; 2018.
2017;14(6):e1002311. doi:10.1371/journal.pmed.1002311. 32. Bell BA, Ene M, Smiley W, Schoeneberger JA. A multilevel
16. Choi SE, Seligman HK, Basu S. Cost effectiveness of subsidizing model primer using SAS® PROC MIXED. In: SAS Global Forum
fruit and vegetable purchases through the supplemental nutrition 2013; 2013.
assistance program. Am J Prev Med. 2017;52(5):e147-e155. 33. Park S, Lake ET. Multilevel modeling of a clustered continuous
17. Fair Food Network. Food Insecurity Nutrition Incentive (FINI) outcome: nurses’ work hours and burnout. Nurs Res. 2005;54(6):
grant program. https://fairfoodnetwork.org/wp-content/uploads/ 406-413. http://www.ncbi.nlm.nih.gov/pubmed/16317362.
2016/09/Food-Insecurity-Nutrition-Incentive-FINI_Overview_ Accessed April 10, 2018.
Fall-2014.pdf. 2014. Accessed May 1, 2018. 34. Benoit K. Linear regression models with logarithmic transforma-
18. Freedman DA, Vaudrin N, Schneider C, et al. Systematic review tions. London Sch Econ. 2011:1-8. http://www.kenbenoit.net/
of factors influencing farmers’ market use overall and among courses/ME104/logmodels2.pdf.
low-income populations. J Acad Nutr Diet. 2016;116(7): 35. Olsho LEW, Klerman JA, Bartlett SH, Logan CW. Rebates to
1136-1155. doi:10.1016/j.jand.2016.02.010.
Incentivize healthy nutrition choices in the supplemental nutrition
19. Freedman DA, Flocke S, Shon EJ, et al. Farmers’ market use
assistance program. Am J Prev Med. 2017;52(2S2):S161-S170.
patterns among supplemental nutrition assistance program recipi-
doi:10.1016/j.amepre.2016.08.023.
ents with high access to farmers’ markets. J Nutr Educ Behav.
36. Patsch AJ, Smith JH, Liebert ML, Behrens TK, Charles T.
2017;49(5):397-404. doi:10.1016/j.jneb.2017.01.007.
Improving healthy eating and the bottom line: impact of a price
20. Stewart H, Dong D. The Relationship between Patronizing
incentive program in 2 hospital cafeterias. Am J Heal Promot.
Direct-to-Consumer Outlets and a Household’s Demand for
2016;30(6):425-432. doi:10.1177/0890117116658237.
Fruits and Vegetables. Washington, DC: US Department of Agri-
37. Cavanagh M, Jurkowski J, Bozlak C, Hastings J, Klein A. Veggie
culture, Economic Research Service; 2018.
Rx: an outcome evaluation of a healthy food incentive
21. Freedman DA, Choi SK, Hurley T, Anadu E, Hébert JR. A farm-
programme. Public Health Nutr. 2017;20(14):2636-2641.
ers’ market at a federally qualified health center improves fruit
doi:10.1017/S1368980016002081.
and vegetable intake among low-income diabetics. Prev Med
38. Baronberg S, Dunn L, Nonas C, Dannefer R, Sacks R. The Impact
(Baltim). 2013;56(5):288-292. doi:10.1016/j.ypmed.2013.01.018.
of New York City’s Health bucks program on electronic benefit
22. Cohen AJ, Lachance LL, Richardson CR, et al. “Doubling Up” on
produce at Detroit farmers markets: patterns and correlates of use transfer spending at farmers markets, 2006–2009. Prev Chronic
of a healthy food incentive. Am J Prev Med. 2018;54(2):181-189. Dis. 2013;10(E166). doi:10.5888/pcd10.130113.
23. Lindsay S, Lambert J, Penn T, et al. Monetary matched incentives 39. Young CR, Aquilante JL, Solomon S, et al. Improving fruit and
to encourage the purchase of fresh fruits and vegetables at farmers vegetable consumption among low-income customers at farmers
markets in underserved communities. Prev Chronic Dis. 2013;10: markets: philly food bucks, Philadelphia, Pennsylvania, 2011.
E188-E188. doi:10.5888/pcd10.130124. Prev Chronic Dis. 2013;10:120356. doi:10.5888/pcd10.120356.
24. Freedman DA, Hunt AR, Merritt K, Shon EJ, Pike SN. Dissemi- 40. Leung CW, Musicus AA, Willett WC, Rimm EB. Improving the
nation of technology to evaluate healthy food incentive programs. nutritional impact of the supplemental nutrition assistance pro-
Am J Prev Med. 2017;52(3). doi:10.1016/j.amepre.2016.08.036. gram. Am J Prev Med. 2017;52(2):S193-S198. doi:10.1016/
25. United States Census Bureau. ZIP Code Tabulation Areas j.amepre.2016.07.024.
(ZCTAs). https://www.census.gov/geo/reference/zctas.html. 41. Kellegrew K, Powers A, Struempler B, et al. Evaluating barriers
2018. Accessed April 23, 2018. to SNAP/EBT acceptance in farmers markets: a survey of farm-
26. United States Census Bureau. American Community Survey ers. J Agric Food Syst Community Dev. 2018;8(81):133-146. doi:
(ACS): 2012–2016 5-year Estimates. https://www.census.gov/pro 10.5304/jafscd.2018.081.010.
grams-surveys/acs/. 2017. Accessed May 21, 2018. 42. Alia K, Freedman D, Brandt H, Browne T. Identifying emergent
27. John Snow I . ZCTA to ZIPCode Crosswalk. https://www.udsmapper. social networks at a federally qualified health center-based farm-
org/zcta-crosswalk.cfm. Published 2016. Accessed May 2, 2018. ers’ market. Am J Community Psychol. 2014;53(3/4):335-345.
28. United States Department of Agriculture. 2010 Rural-Urban 43. Kahneman D.Maps of bounded rationality: psychology for beha-
Commuting Area (RUCA) Codes. https://www.ers.usda.gov/ vioral economics. Am Econ Rev. 2003;93(5):1449-1475.
data-products/rural-urban-commuting-area-codes/documenta 44. 115th U.S Congress. Agricultural Improvement Act of 2018.
tion/. 2010. Accessed May 2, 2018. 2018.